| Title |
Dirbtinio intelekto metodai radiologiniais skaitmeniniais vaizdais grįstų klausimų atsakymams prognozuoti / |
| Translation of Title |
Artificial intelligence methods for predicting questions answers based on digital radiology images. |
| Authors |
Gaidamavičius, Dainius |
| Full Text |
|
| Pages |
64 |
| Keywords [eng] |
deep ; neural ; networks ; radiology ; images |
| Abstract [eng] |
According to World’s Health Organization data more than 45% world countries have less than one physician for thousand residents. [1] Because of this reason medical staff are forced to work large amount of hours and this situation leads to growing human error rate. Since physician’s work requires a lot of time to analyze medical images deep learning could come to hand and automize image processing. Nowadays, machine learning researchers invest more and more attention into visual question answering problems and these endeavors lead to models which can understand wider context in given image and solve problems which requires human level reasoning. In this work were analyzed deep learning methods which goal was to predict answers about digital radiography images. In this work were applied ResNet50V2, EfficientNetB0 convolutional neural networks, long-short term memory, recurrent neural neworks and proposed new image and text feature vectors fushion options. Experiments have shown that best architecture used feature fushion by adding feature vectors into custom matrix and further applying convolution layers. Highest mean average value was reached by ResNet50V2 and LSTM networks and was equal to 84,80%. |
| Dissertation Institution |
Kauno technologijos universitetas. |
| Type |
Master thesis |
| Language |
Lithuanian |
| Publication date |
2022 |
